Port JSONCompleter and drop the PartialJSONDecoder dependency - #259
Merged
Merged
Conversation
GeneratedContent(json:) completed truncated JSON by appending a single closing brace, bracket, or empty string. That failed on every realistic streaming cut, including nested objects, open arrays, partial literals, and the example in its own doc comment, and turned a trailing backslash into a literal quote. The system model separately pulled in PartialJSONDecoder to get real completion for the same job. Port JSONCompleter from PartialJSONDecoder as an internal type and use it in GeneratedContent(json:). It walks the text as a JSON value and appends exactly the closing characters the input is missing, with a depth limit. Text that ends inside an escape sequence is cut back to the escape start so the result stays valid JSON. The system model now calls GeneratedContent(json:) directly, and the package dependency is removed.
…mmas A complete empty object or array was reported as needing a closing character, so the enclosing container closed right after it and every later value was discarded. Whitespace between a complete value and the following comma was treated as a missing comma with the same effect. Treat an empty container as complete and skip whitespace after a value before looking for the comma. Also cite the exact upstream revision in the attribution: the ported source is MIT licensed at that revision, and the same source ships in the 1.0.0 release under Apache-2.0.
findEndOfCompleteValue re-ran the completer on a value its callers had already found complete, then scanned it again to locate its end. Each nested complete sibling doubled the work, so a complete sibling nested 20 deep before an incomplete tail took half a second and 30 deep would take minutes. This runs on every streamed snapshot. Drop the rescan and rely on the linear string, bracket, literal, and number scanners. The depth counter also advanced twice per nesting level, once entering a container and once per element, so the effective limit was half the documented value. Count each level once.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
GeneratedContent(json:)completed truncated JSON by appending a single closing brace, bracket, or empty string. That fails on every realistic streaming cut, including nested objects, open arrays, partial literals, and the example in its own doc comment, and it turned a trailing backslash into a literal quote. MeanwhileSystemLanguageModelpulled in PartialJSONDecoder for real completion of the same kind of input.This ports
JSONCompleterfrom that package as an internal type and uses it inGeneratedContent(json:), so truncated JSON now completes to the exact structure the model was emitting. Text that ends inside an escape sequence is cut back to the escape start so the result stays valid JSON.SystemLanguageModelnow callsGeneratedContent(json:)directly, and the package dependency is removed.